What jobs are AI jobs?

Hi, I'm Tony Cameron Brown. And I'm Benedict Evans. That hasn't changed. That's good. Yeah, not that. Yeah, a week to get a month to go. The thing I thought it was all I was talking about was, tell me, trying to predict what jobs, what industries, what companies will be affected by AI. Because anthropic does this thing, and openly I have done these things based on sort of US census data where they try and put like a numeric score job by job of exposure to AI, which seems to me just absolutely due to Chris as an exercise instead of self-deception that like you can't possibly know this. And then there's a bunch of other people kind of proposing different kind of two-by-two frameworks like Sequoia did one and Kossa did one in the last couple of weeks. And it's struck me it's just kind of interesting to talk about like what you can know what you can't know. And in particular to say, well imagine you've been doing this in 1997 with the internet, what would you have got and what would you not have got. And the funny thing about that is like we now forget what wasn't obvious back then. It's so obvious to us now. And actually before we dive into that question for you that I was thinking as you were saying that, did we fret so much about what jobs were going to be replaced by technology whether that was the where, but the mobile, back in the day, the same way that we are fretting so much about. Yeah, it's interesting. No, because there was there was less the the first obvious thing that you can do with this is automating existing jobs. Which wasn't quite so much the case with with with the web. It wasn't like the first obvious thing is that this will automate X and Y and Z. So it's the automation piece that has ever because I have don't feel like I've ever heard so many people focused so heavily on the fact that these jobs or their job or the neighbor's job is going to get completely erased. Whether it's right or wrong. Yeah, because it's just a different character of the technology. Although you know you go back and look that is kind of what happened with the internet and with PCs just not the same way. And so the thing I was sort of thinking was like you know as I started my career as an analyst in 1999, Spring of March 9, justice said like the bubble was really starting to go to inflate. And so you would look at e-commerce. You know imagine your clever and hardworking and you know you'd look at what I am, you know, you know, you what would you do? Well, you look at retail and you'd say I mean the framework that went around was what's high touch and low touch. This is a sensible way of looking at it. And that would tell you that like consumer electronics would go much more quickly than high fashion, which was more or less true. There were some interesting wrinkles within that. So like it was not clear that like makeup would be easy to do online, for example. And then you would kind of. And then finally enough this stuff would like close that we're still struggling today with return and exchanges and women buying clothes that actually fits them. We're just fascinating. 30 years later we're still struggling. Here we are 25 years later and people are still trying to work out how this works. And 30 years later and people are still trying to work out how this works. So that's one piece. And then you would kind of go and look at industries and you would say well who is about information arbitrage? And where are the physical assets that used to matter and don't matter anymore? And so you would look at that and you would say you know and then and then then the interesting thing is to kind of go and back test it so the thing I've point I've made in meetings recently is to say well imagine you were doing travel. So you would say well obviously travel agents have got a serious problem. And if you're a hotel or an airline then you need to think about price equititry and price transparency and maybe you need to think about loyalty programs. But in the end your business is about owning a physical asset. And that was true and then Airbnb came along 15 years later and said no maybe it isn't about owning a physical asset. The same thing for um taxis you would have not nobody nobody realized that Uber was coming and even when it happened nobody understood that this made any kind of sense and people still kind of sometimes argue about it. So there's a sort of an Uber test to all of these frameworks which was what happens how much are you like doing linear extrapolation from the current market how much are you just presuming that you'll use this new thing to do the old thing but better. Yeah. There was a valuation professor who notoriously did an analysis of Uber kind of 10 years ago where he said you know the valuation of Uber should be based on the fact that its term is a size of a taxi market and this is I'm going to work out the size of the taxi market and therefore we'll get extra cent and therefore that's that's what Uber's worth and of course this was completely wrong because Uber's term was not the taxi market it was something else it was much bigger than the taxi market and so you get this sort of sense of well you can do the linear extrapolation and you can have your framework and that framework will generally be directionally correct that every now and then someone will come along and pull the whole thing inside out and say I'm going to make it I'm suddenly your framework is going to be the wrong way of looking at the question and you know go back and read look at retail like most of the time it's right. Trying to put a numeric score or is it is idiotic you know people now trying to say well you know a countenance of God exposure of 72 and bookkeepers have got an exposure of 84 this is just insane you know it's like there's an old joke about the physicist who asked to predict which horse is going to win a horse race and the physicist says first we presume that each horse is a perfect sphere but you're trying to simplify and to get to a degree of precision that's impossible. It is crazy as humans that we like a good number regardless that makes us feel good and though it could be completely wrong but we're just like oh but that 72% helps me wrap my head around it does it even if it's complete bullshit okay. Yeah I mean I'm subject to this too you know sometimes I make a slide where I'm making kind of making a tropical point and I'm doing it with a chart but we all kind of know the point the chart is just illustrating it even and the numbers aren't that important. So this is kind of the thing that you can come up with these frameworks so you can say you know is AI more about information retrieval or is it more about intelligence is it about knowledge or is it about intelligence or this is your point about reframing the questions we thinking. Yes every now and then and those things are always kind of useful but there was going to be but there's if there's going to be one that's wrong and the thing that I was sort of thinking about as maybe getting to a high level of abstraction here is to say you know the Clay Christensen phrase what is the job to be done what is the actual thing that the customer is buying from you and how do you map that against what is the thing that this new technology is changing and so you can look at and where is the point of leverage and so simple example would be you know airlines okay the online flight booking is all going to change and that's going to change a bunch of stuff about your pricing and maybe some of your margins but in the end the booking isn't the product the product is a plane that takes you from A to B on back and the plane isn't being changed by this thing. On the other hand if you think about retail say maybe there's a bigger continuum here so there are some retailers where the business is to be the most efficient endpoint to a logistics system and if internet shipping is a more efficient endpoint to the logistics system then you're out of business. There are some cases where internet is not more efficient like grocery which is why Walmart is still a giant business because it's actually more efficient to drive to the supermarket than it is to deliver all of that product with the cold chain and the stuff that brews is and everything else and deliver that to everybody's home and you know here we are Amazon is still you know poking away at it after 25-30 years but your grocery is different but for everything else if your retail proposition is to be the end point to a logistics chain the internet would do it better. It wasn't clear that Amazon would do all of that and not just books in 1997 but it was clear that like the internet was going to be this tough fundamentally new end point to a logistics chain. On the other hand if you were a retailer his proposition was not being the efficient end point to a logistics chain but something else like if it like experience, service, digestion, curation the word everyone suddenly started using his taste. If it's about opinion rather than we have every widget then the internet wasn't much of an issue which is what happened is not really good at the extreme case but also fascinating again book, bookstores. Some of book purchasing is I want that book now and I know what book I want but a whole other part of book purchasing is I don't know what I want and book shopping as a leisure activity. I'm going to go into the book shop for an hour and I'll walk out with three books I didn't know it is. And I want recommendation and I want to your point someone with taste to understand what I'm looking for and that can give me good advice. And the recommendation in a sense is even if it's just that you've only got one for five thousand books that is a constraint that drives recommendation where if Amazon has seven eight nine hundred Amazon has seven eight or nine hundred million schools so they can't do recommendation. And so that's the kind of question what's the point of leverage? And newspapers I think are kind of an interesting midpoint in that you know you can make a joke which is maybe even be true that newspapers are rich at the internet and thought this is going to be great because our printing bills will go down. And but you know to take that point kind of seriously you know what is a newspaper? Well you are everything encompassed by the word journalism like recommendation and suggestion and curation and investigation and journalism and then you are trucking a light manufacturing company. And clearly the internet means you won't need the trucking a light manufacturing but that's not what we're buying. Customer isn't buying pieces of paper for you. They're buying the journalism from you. So the first run at this framing would be well well but we are fine just as Chanel is fine because that's not what people are buying people are buying the store they're buying the product and now we can deliver that product in a different way. The problem was the physical asset was also your barituary and it was also what protected you. So you've got this kind of blit is like you've got you know the internet unsplitts these two things apart you've got this physical asset and you've got the product and where is the point of leverage or maybe they're both points of leverage in different ways. And so those get broken apart and then you don't need one is that better for you or is that a catastrophe for you or did it not make any difference. And I think we saw that from the gentleman. realistic perspective, I mean, you and I have spoken about this before, but with something like a sub-stack where the baritone tree now is any journalist or anyone with an opinion and a following can create a newsletter that because the baritone tree is now so low and anyone can do it, well I found myself finding myself, the situation that I found myself with is overwhelmed by the amount of people who are putting things out there in the world and what I really need is someone to tell me, okay there are 20 people writing about LLM, these are the two people of substance who actually have an opinion that's founded in experience or expertise and Benedict. I wasn't joking, it's so special. But what is shifting is interesting, what we initially thought, oh this is, and we saw it during the pandemic, the amount of journalists who are just like, I'm out, I don't know what one's going to be associated with the big media entity, I want my freedom, I and now the baritone tree from the tech perspective is so low and then realizing, oh it's actually hard getting people to give you money for being a single entity and being a soul writer. Yeah, well the other case here would be our publisher, so it was his brief moment when people thought that also we just go direct and of course that isn't what happened at all. So anyway, so the thing that the point of the framing is, like you've got, you know, generalise it from the internet, you are doing this thing that customers are paying you for and then there's a thing that you have to do in order to deliver that, which is having a store or printing paper or something and maybe doing that thing was actually what people were paying, maybe the thing that the internet automates, it was actually what your business was and it's gone or maybe that was the baritone tree and you have a problem or maybe actually that wasn't really anything to do with what your actual business was and so not life changes, which would be the case of the airline. Well, maybe it's both. It's the end product and the experience or the ease of the experience or the ease of the delivery method and you're just like the sum total of those two is worth more. Or maybe even better. Now, but the point is you've got those two separate things. You've got the actual thing people are coming to you for and then you've got the way you're doing it. Are those separate? If they are separate, what happens if the way you're doing it, get automated away? Is that much better for you or does that create new into me doing these, like the new OTAs or is it much worse for you? Newspapers. It seems to me you can apply exactly the same frame into AI, which is to say like, okay, you are a both of them. Now you can make the document as much quicker. Is that why people were coming to you? Is that what people were paying for? Now you are a software development company. You're a software company. You make vertical SaaS. Now people can write the code much quicker and easier. Okay, it was at the product with what people is the reason that XYZ SaaS company doesn't have competition or has little competition that you'd have to write this many hundreds of thousands of lines of code to do it. Now I was at Andreessen Horace for six years. You were, I didn't know, five or six years, a nation builder. I don't feel like the reason it's hard to compete with X or Y or Z company is how long it would take you to write the code to replicate the product. It's almost never the problem. The problem with the hard part is everything else. It's working out what the product should actually be and what the code should be doing. And how you're going to insert this into the industry and persuade people to use it and executing the Salesforce and getting the right go to market and the right pricing and working out how you coexist with people in the industry already in the industry and who you don't compete with and who you do. It's all the other stuff. It's not writing the code. You've got to turn your notifications off. You're giving me bubbles that sums up. Now of course, you know, that said, you know, there are, I'm sure there are software companies who are writing the code. It's a hard part. They give argument like if your, if your business is maintaining cobalt code from regional banks in the USA, then yes, you've got an existential problem there. Not nice yet, but you know, clearly this is going to fundamentally change your business. But if you are door dash, the hard part of door dash is not writing the code. This was the astonine part of that email that went around a couple of weeks ago. Like, oh, you'll just get the agent to write the code and then the agent will just go and sign up the retailers and they have the restaurants and the drivers. Like, tell me you name nothing about how the places work. The point is that's not the hard part. The code isn't the hard part. And sometimes the code is a hard part. The same thing with professional services. Which part of what you go to to a lawyer for or an accountant for is a thing that's not going to be automated, which isn't. And the thing I always go to do to a pair of charts here is what you have to chart. The first pair of charts is, starting out, I've often talked about made a centerpiece of my last presentation, which is elevator attendance. Used to be that elevators were manually operated. They have an acceleration of break and there were like a hundred thousand people in America employed as elevator attendance. And then that gets automated. It turns into button that you press and you don't need an elevator attendant anymore. So the data person's job is to be a button and that job is automated. And nothing except in a few apartment buildings in Manhattan where it continues as an amenity. The other side is look at accountants. The number of accountants in the USA increased every single decade in the 20th century. Add you go through adding machines and computers and mainframes and PCs and Excel and ERPs and SAS. You get wave-off to wave of automation and number of accountants keeps going up. It should tell you that the job wasn't adding up the numbers and putting them on a piece of paper. It was something else. It wasn't counting the numbers and making sure they matched the numbers on the other piece of paper. That wasn't the job. That was just how you were doing it. And there's a sort of, I'm sort of thinking out loud here, but there's a sort of a pricey elasticity question in here. And it's another slide that I made last year and I'm using again now, which is, if you make it cheaper and easier to do something, do you do that more or do you do that, do you do the same amount for less money or do you do more for the same amount of money? If you make it cheaper and easier to build spreadsheets, do you do more spreadsheets? Or do you do the same amount of spreadsheets for fewer people? But there's a step two is, is having to have all the people to do that, your barrier to entry, which is, and both of these points apply to media because we made it much cheaper and easier to distribute content. So the result is that we have massively more content. And the barrier to doing, and needing to have the physical infrastructure with your barrier to entry, so that's not a way. So you have massively said in these papers, a completely screwed, magazines are screwed, there's massively more stuff more content created. But there's a step three, which is what are the things that you couldn't do because the only way to do that would be to have like a million people on paper, on paper. And these, the example, like the extreme example I gave, we're sorry, I'm on logging, like the extreme example I gave, was like, if you wanted to have an express train that went from London to Scotland in 1800, it wouldn't matter how many horses you bought, you still couldn't do it. You could put 10,000 horses on the front of this thing and it still wouldn't be able to go from London to Scotland in eight hours. You could not want to want to find if you had a million people working for you in no computers, you just like that wouldn't work. And so there's like the automation and then there's the barrier to entry and then there's what just wasn't possible without this before. And I also think there's another piece here, which is what are we putting, how to say this? It used to be that we got really excited and impressed by the technology that was available and then it felt like, and I already felt this like 10, 15 years ago, and then the human element was sort of less important. And I think I saw this when we were selling SaaS. It was like, who had the better tech and who cares about the salesperson or the implementation team that you had? And it feels almost like there's a shift now that our focus and what we're more excited by is actually what does the human element and human implementation of all of this look like? Because we're less impressed by the tech because we're realising that the tech that we're going to be able to be to build is unfathomable. We know that we can create incredible tech. And so I think what I'm getting more excited about is how are the humans going to implement this tech? And we saw this in SaaS when I remember having this tension with our founder where he was like, we sell software, we don't sell human services. And I was just like, yes, but you need the human services, consultancy element. If you want to close the $100 million deal. Someone's going to pay $100 million just for the technology that you can buy off the shelf. They need to understand how they can map this to the needs that they have. That is the consultancy element. People still want to sit in a room with someone who's incredibly smart and who can tell you, I worked on a Balmer's campaign. This is what worked for us. Yes, here's you've bought the right tech. But here's what we've learnt and how we think we can reproduce this for a European campaign, for example. And I think that I don't know, I feel like we're less and less impressed by the tech and we're going to be more impressed by what the humans can do with it. Or maybe that's completely wrong. It's funny, I'm imagining the Drake meme of No and Yes. And the Drake is saying, professional service is something no. Forward deployed engineers, yes. But it almost feels like we're so hard to comprehend today what the tech is capable of doing, partly because of everything that you've talked about. We just don't know yet. I think I'm, and I had this argument with my husband and I was asking him something. He's like, have you asked Claude, chat GPT and I'm like, oh, for God's sake, I actually want to have a human conversation with my husband about something. I don't want to sit in front of my laptop. The reason that you get married is to problem solve together. So the fact that that is being referred to me now, just like, have you asked chat GPT? And I'm like, oh, my God, no, what are we doing here? This is, I'm not like, have you asked up to Google yet? But so there's something there. I'm just like, I actually, I don't know. I wonder if I'll point of leverage or or what we're going to focus on more. We've talked about this before like Netaporte and just their extreme, you know, their approach to e-commerce is very much a white-loved experience. They can get here as fast as an Amazon delivery, but you're paying for something more than that. I don't know. I just feel like I'm having a hard time wrapping my head up. This is the key of this next generation of tools and technology. Well, some of this is also, I mean, there's a, there's a price-y elasticity point, I saw someone say, like I think in the dot-com bubble, like somebody said this point, there's a big difference between making something very cheap and making it free. There's a sort of a binary difference between making it very cheap and making it, when you make it free, you unlock all sorts of stuff that you didn't have when it was just cheaper. I think there's another bit here, which is something I think I'm going to react to thinking about. It is LVMH and the other luxury gaskin glommets and where you're selling mass produced, mass retailed, mass manufactured, mass merged, and biased individuality and uniqueness. But you're also selling people something other than this is just the thing that's in the store. It's about the experience and the curation and the taste, you know, the word at the moment. But there's, there's, I mean, that's a specific set of cases where that might be the difference or that might be the point of the leverage. But I think the real question here is, like, you're going to automate this thing now. And that will vary the impact of that will vary a great deal by industry, by company. And it will vary in like weird, unpredictable ways. So it will be stuff where it didn't occur to anybody that you could automate that thing. And now you can, we didn't occur to anybody that people would want that automated. And now you can. And this is part of the story, you know, the story of Amazon in the last 25 years is on the one hand converting stuff that people thought needed high touch into stuff that you turned out you didn't. I mean, it's like, you know, I'm trying to buy a pair of shoes right now and I bought three sizes from Mr Porter, rather than hunting around who in New York stocks is particular brand and has this and saying, I'm just, you know, how are that? Just buy three sizes and they'll live in the course of three or four days from different retailers. And I'll return the two, the two that I don't like. And so we've had this process of converting high touch into low touch. And what we had with the internet was, you know, it had out to that, you know, Instagram substitutes for celebrity magazines. And Instagram is North American magazine. It's not even journalism. It's some whole other thing that substitutes for, you know, it's like you're kind of pushing down massage hierarchy to finding low and level, low and level, different level of abstraction of how you would think about what this is, which is why, at what I said at the beginning, you think you've made this great framework and then there will be something on your framework where it turns out that it's just not that at all and someone turns it into the opposite thing. But the framework to think about is, okay, we've got this automation thing here. It's the thing that it can automate the thing that you do. Or is it just the way that you do the thing that you do and what you're actually selling is something else? Sorry, that sounds like that top-hand maybe. Sorry, that was not deliberate. But what is it that people are actually coming to you for? And is this automation going to change that or is this automation sort of incidental to that? And as I said, there's two parts to that because it may be that you are, you know, it may be that you are the bit that's getting automated three parts. It may be you are the thing that's being automated. You are the electronic store. It may be that you are the newspaper where that isn't really what people were buying from, but that was a baritone tree. And it may be that you are the airline where in the end that whole part of the value chain was kind of incidental and credit cards were a much bigger deal. It's funny, I spent some time on this a couple of weeks ago trying to make a chart of audit costs. You know, what I came up with was a chart of how many people were employed as a character since 1900 and the number, it's done up every decade. But I was wondering is there data on what the average company pays for an audit? In terms of how it is there is and it hasn't changed since about 2003. And then there's data, I've had other studies that went back to the 80s, but the the interesting part of it is that you get these 30 page academic studies on audit costs. And they talk about all sorts of stuff and they never mentioned computers. There's all these other stuff that's going on in the industry that's changing audit costs. And this is actually only one or five or 10 things that's on your list. I remember when I was at Andrews and Horowitz, I remember having a meeting with a big utility company, a European utility company. And the CEO listened to my presentation about what we then called AI, which is now just machine learning. And he says, this is very interesting, very interesting. I think maybe innovation will be one of our top five priorities next year, which if you're in the valley, sounds completely insane. And then you think, but this is a water company, like maybe they probably do have other priorities. Like there's leather and pipes and there's an earthquake and a drought and regulation and like they're worried about the Russians blowing stuff up. And like there's all sorts of other stuff. But there's more important than than innovation if you're a water company. Like you dig holes in the ground and they're there for 50 years. And I had a conversation with somebody who was then became ahead of one of the big global ad networks. It's like 2016, 2017, like what do you worry about? And he said, well, obviously, and I'm worried about AI and I'm worried about social and Google and so on. I'm also worried that one of our countries doesn't have a good head of creative. It was like with all the other stuff. It's like the line, you know, what do you do when in an emergency flying an airplane, first of all, fly the airplane, all the other stuff. And by the way, I'm seeing this first hand with a company like Apple, who's just got the broadcast media rights for Formula One. Their first priority is can, you know, before all the innovation, before all looking at all of the original content is can this succeed? Can we make sure that the product doesn't change and that the fans tuning in can actually see the race in decent quality? And so it's that short term gain versus the long term, you know, stuff that they want to work on. And it is fascinating to your point, but it's also interesting to see what fans are looking for versus what the consumer sorry is looking for. And you're never going to satisfy them at the same time. If the broadcasting wasn't as good as what it previously was, they would have gone up in arms. And at the same time, these same consumers are going, oh, it's so nothing has changed. So Apple hasn't innovated. And you're like, we're on race two. What? Yeah, exactly. What are you going for me lately? Because again, if they hadn't to focus on the immediacy of just to your point, what are the priorities here? Well, the priorities are making sure that fans tune in and don't see a difference. Then we want them to be wowed and see a difference. There's a, I mean, I can't remember who it is that said that all all model of the wrong but all useful. And it kind of goes back to that point. I mean, you know, you could, you know, this is, you know, open AI and topic model that like we're going to score jobs. We're going to deep duty, evil. We're going to score jobs by exposure to AI. My fiance was looking at this and she was saying right, but accountants have been ordered to be people have been ordered. This says that accountants have got really high exposure. We've been auditing or automating accountants for 125, 150 years. And there's more every year. So you might want to revisit your assumptions here. And meanwhile, you've also said that fitness instructors are going to be really, really safe. And have you thought about what happens if I put my lean my phone again, put my phone on the chair and pointed out me while I do my exercises and connect to an AI agent that analyzes what I'm doing? It may be that they'll be, this is the Uber example, the Uber test. There'll be a bunch of things that you think your framework tells you are completely safe and it turns out someone will work out a way to turn that into into automation. And meanwhile, there'll be software like people have been trying to automate that thing for 150 years and it doesn't work. And so those framework, that framework in particular, the thing that really drives me crazy about that framework is the numeric accuracy of it. It's like we're going to give a percentage score that you have no idea. You can, most you can do is say, probably higher, probably lower. But there is this is more general point of like, it's not physicality that's the question, you have to go, maybe this is what I've been broken towards as I've been rambling, I'm not longing for the last half hour. It's like the question isn't physicality is your product physical or not? The question is does physicality matter? It's not how easy is it to automate that job with a computer. It's this thing that you automate the actual job. You know, I mean, I swear to God, I look at these scoring systems, it's like, well, I'll always spend 32% of their day on phone calls and AI can do voice therefore AI can automate that. The man, that's not the maths. Yeah. That's not what the job is. That's not the job. I like that. That's not the actual question. And that's the question. We're realizing we actually don't know what the job of most people is to do. What do they actually pay? Which comes to, you know, it comes to them, the McKinsey thing. What do you pay McKinsey for? It's not the slides. It's it's other things. Sometimes it is. Sometimes you pay, you bought, you paid for a private equity, do diligence deal. Sometimes you pay for them to validate this so the strategy you've already decided on and they give you a bunch of slides that tell you what you already knew. But that's not what you do as a part, not a senior partner at B&B, CG McKinsey all day. That's not the job. And to your point, everyone is providing the same software or the same type of PowerPoint decks. What's the differentiator here? And the differentiator most of the time is the human that you have in front of you. Well, this is also the question is, you know, typical big company today has four to five hundred and a SaaS company, SaaS apps. And all those are is a bit of business logic wrapped around some sequel. And so why is it that at Zelo and Airbnb and Tinder all exist and they're not, and there aren't 50 of each of them. Well, because it's not just a bit of sequel and some business logic and you are you something else. There we are. That's a good place. It's harder than you. It's harder than it looks. Yes, it is. Another good conversation. See you next week. Good to chat. Bye. See you next week. Bye.